NYPTIDMeridian
NYPTID Meridian · Agent workspace
Every frontier model on one subscription. Agent runs that survive a crash, check their own work in a real browser, and wait for your sign-off before anything counts as done.
ADE from $80/month · No API keys needed
Windows desktop app and web. Echo voice sold separately or bundled.
Not a screenshot. Filter the dashboard, open an agent, and mark the finished run verified.
One subscription. Every frontier model, switchable mid-session.
Anthropic Claude, OpenAI GPT, xAI Grok, Google Gemini, Moonshot Kimi, DeepSeek, Qwen, Meta Llama, Mistral and Z.ai GLM.
Why Meridian
Most agents are impressive for ten minutes. These are the pieces that make them useful for a whole day.
Durable runs
Every step of an agent run is written to a journal before it happens. If the host restarts, the run comes back paused with its history intact, and resumes from the exact step, never repeating a write it already made.
It checks its own work
Meridian agents can load a page they just changed in a real headless browser and read the screenshot, or hit the endpoint and read the response, before they tell you it works.
Review gate
A run finishes by proposing a result with evidence. It sits at "Ready for review" until you mark it verified. A confident paragraph never counts as success.
Agents dashboard
See which agents are working, which finished, and which are waiting on you, with how long each has been at it. Click one to bring it into a pane.
Memory & skills
Agents propose facts and reusable procedures as they work. Nothing enters their context until you approve it, and skills are versioned as they improve.
Routines
Put an objective on a schedule. Meridian runs it as a full agent run and leaves the result on your board for review. A routine never stacks on its own unfinished run.
Connections
Agents can see the tools a connection offers, but every call out to the real world stops and asks you first, showing exactly what it wants to send.
Also from Meridian · Echo
Echo is system-wide voice for Windows. Hold a shortcut, speak, and the words land in whatever app is focused: your editor, a chat, or a Meridian prompt. It also runs live two-way calls in over 100 languages.
Questions
No. Model usage is billed from your plan's included cloud credits, and you can buy more if you need them. You never paste a provider key into Meridian.
ADE agent runs execute on Meridian's cloud host, which is why they keep going when your PC is asleep. Connect GitHub so agents can work on your repositories there. File writes and every external tool call can be gated behind your approval.
Frontier models from Anthropic, OpenAI, xAI, Google, Moonshot, DeepSeek, Qwen, Meta, Mistral and more. You pick the model per session and can switch mid-session.
A chat answers. A Meridian run works: it reads and edits files, runs commands, checks its own output, and ends with a result and evidence for you to verify. Runs are journaled, so a crash doesn't lose the work.
Echo is Authenticode-signed through Microsoft Artifact Signing. SmartScreen may still prompt until the publisher is widely seen. Match the SHA-256 on the download page before you run the installer.
Pick a plan, open a workspace, and hand it the task you've been putting off.